Nonabstinent alcoholics and social drinkers were presented with an ad-lib supply of either alcoholic or nonalcoholic beverages in a taste-rating task. Subjects were assigned to one of two instructional set conditions in which they were led to expect that the beverage to be rated contained alcohol (vodka and tonic) or consisted only of tonic. The actual beverage administered consisted of either vodka and tonic or tonic only. The results showed that instructional set was a significant determinant of the amount of beverage consumed and posttask estimates of the alcoholic content of the drinks. The actual beverage administered did not significantly affect the drinking rates of cither alcoholics or social drinkers. Loss of control drinking, in the form of increased consumption by alcoholics who were administered alcohol, did not occur during the drinking task. The results are discussed in terms of implications for treatment and for the conception of alcoholism as a disease.